This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2018.2819162, IEEE
Access

0.5, cnum =  2*0.5 = 1. For C1, one bit (either the All particles update their velocities and positions
repeatedly, until the best solution is found or the maximum
second or third) will be selected for crossover. Suppose the number of iterations is reached.
second bit is selected to be changed from 1 to 0. Then, the
new C1 is 10010. Next, suppose the fifth bit is randomly B. ALGORITHM DESCRIPTION
selected for mutation. After changing the fifth bit from 0 to 1,
Based on Bio-HUIF, Algorithm 5 describes the proposed
C1 becomes 10011, representing itemset bef. This is a PEV,
Bio-HUIF-PSO algorithm for mining HUIs.
contained by transactions T'2 and T'10, and u(bef) = 131 >
Algorithm 5 Bio-HUIF-PSO
min_util, SHUI={bde: 166, bef: 131}. We can also obtain the
Input Transaction database D, minimum utility
child of C2 using the same routine. Suppose the new C2 is
value min_util, maximum number of
10000, representing itemset b. As u(b) = 216 > min_util,
iterations max_iter
SHUI={bde: 166, bef: 131, b: 216}. After generating these
Output HUIs
two new chromosomes, SNnew = 2. As SNnew < SN, two
chromosomes of the current population will be selected and 1 Pop_Init( );
the above process of crossover and mutation will repeat until 2 times=1;
3 gbest=;
SNnew  SN.
4 SHUI=;
Suppose that no HUIs are generated by the other child
5 while times < max_iter do
chromosomes of C3 and C4, and SHUI is still {bde: 166, bef:
6 for i=1 to SN do
131, b: 216}. According to Algorithm 3, two HUIs will be
7 if times==1 then
selected using (7). That is, b has the highest probability of
8 pbesti=Pi;
being selected and bef has the lowest probability of being 9 end if
selected. After representing the two selected HUIs as bit 10 X = IS(Pi);
vectors CNi and CNj, two randomly selected chromosomes 11 if u(X)  min_util and XSHUI then
from the second population will be replaced by CNi and CNj. 12 XSHUI;
The above process is then repeated for the new population 13 end if
until the termination condition is satisfied. 14 if u(X) > u(pbesti) then
15 pbesti = Pi;
V. BIO-HUIF-PSO 16 end if
In this section, the PSO algorithm is briefly discussed before 17 end for
the proposed Bio-HUIF-PSO algorithm is explained. Finally, 18 Find gbest among SN particles;
an example is used to illustrate the application of Bio-HUIF- 19 Next_Gen_PA( );
PSO. 20 Determine gbest using roulette wheel
selection among HUIs in SHUI;
A. BASIC IDEA OF PSO 21 times++;
PSO is a bio-inspired algorithm that simulates the foraging 22 end while
behavior of birds or fish [16]. 23 Output all HUIs.
In the PSO algorithm, several particles are initialized at In Algorithm 5, the procedure Pop_Init is called in Step 1.
random. Each particle moves toward the optimal value Step 2 then initializes the iteration number to 1. In Step 3, the
according to the following two equations: global best particle gbest is initialized as the empty set. The
set of all HUIs SHUI is also initialized as the empty set in
vit 1  wvit  c1r1 ( pbesti  xit )  c2 r2 ( gbest  xit ) (9) Step 4. The main loop (Steps 5–22) discovers HUIs
population by population. The SN particles are checked one
xit 1  xit  vit 1 (10) by one in the loop from Steps 6–17. If the current population
t t 1 is the first population, the current particle Pi is also initialized
where vi and vi are the velocities of the ith particle at
t t 1 as pbesti (Steps 7–9). Step 10 determines the itemset that
iterations t and t+1, xi and xi are the locations of the ith
corresponds to the enumerating particle. Here, the function
particle at iterations t and t+1, pbesti is the previous best
IS( ) returns itemset X by unifying the items in Pi if its value
location of the ith particle, gbest is the current best location
is 1. If the current particle can produce an HUI X that has not
of all particles, the three constants w, c1, c2 are weighting
already been discovered (Step 11), Step 12 records this
coefficients, and r1, r2 are random numbers in the range (0, 1).
itemset. Steps 14–16 update the pbest of the current particle.
The first part of (9) includes the previous velocity and
In Step 18, gbest is updated by the particle corresponding to
constitutes the momentum component; the second part uses
the discovered HUI with the highest utility value. Using the
the previous best position and constitutes the cognitive
new gbest, Step 19 calls the procedure Next_Gen_PA to
component; the third part takes the best previous position of
generate the next population (described in Algorithm 6). In
the neighborhood and constitutes the social component of the
Step 20, initial gbest of the next population is selected by (7)
iteration.

VOLUME XX, 2018 9

2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.